Heidelberg 2022 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
T: Fachverband Teilchenphysik
T 53: Data Analysis, Information Technology and Artificial Intelligence 3
T 53.8: Vortrag
Dienstag, 22. März 2022, 18:00–18:15, T-H38
Refinement of jet simulation with generative adversarial networks — •Shruthi Janardhan1,2, Sven Harder1, Patrick Connor1, Peter Schleper1, Daniel Ruprecht2, and Sebastian Götschel2 — 1Universität Hamburg — 2Technische Universität Hamburg
In High Energy Physics, the interaction of particles with matter at the detectors are best simulated with the GEANT4 software. Alternatively, less precise but faster simulations are sometimes preferred to reach higher statistical precision. We present recent progress of refinement of fast simulations with ML techniques to enhance the quality of such fast simulations. We demonstrate the use of generative adversarial networks in the context of jet simulation using a Wasserstein loss function. The architecture consists of two opposing networks, Refiner and Critic. The Refiner, refines the distribution of the energy of the jets obtained with the fast simulation. The Critic is used to effectively differentiate between the distributions of refined energy and the distribution obtained by the GEANT4 simulation. The Refiner can be used solely to obtain a fast but refined jet simulation.